A tunnel lithology real-time identification method based on a clustering algorithm

Abstract

This application relates to the field of tunnel construction technology, and in particular to a real-time tunnel lithology identification method based on a clustering algorithm. The method includes: collecting and preprocessing multi-dimensional drilling parameters using a drilling rig to generate a standardized dataset; mapping these parameters to a high-dimensional feature space to form multi-dimensional feature vectors; performing unsupervised classification using an improved K-means++ algorithm to dynamically optimize the centroid distribution and generate the final lithology category; and linking with a geological database to output lithology identification results and guide the adjustment of construction parameters. This application can quickly adapt to lithology changes in complex construction environments, significantly improve the real-time performance and accuracy of lithology identification, reduce the risk of misjudgment, improve construction safety and efficiency, and promote the development of tunnel construction technology.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering and intelligent data analysis technology, and more specifically, to a method for real-time identification of tunnel lithology based on clustering algorithms. Background Technology

[0002] With the continuous advancement of tunnel construction technology, engineering projects involving deep-buried tunnels and complex geological conditions have been widely carried out. However, in these major engineering projects, the construction environment is complex and the rock properties change frequently and in various forms during the excavation process. This situation has a direct impact on drilling and blasting design, support parameter setting, and construction safety. Existing lithology identification technologies have significant limitations, mainly relying on manual experience or geological sampling and analysis. Manual identification is usually based on drilling sounds, equipment vibrations, and characteristics of slag discharge. However, this method is highly subjective, lacks stability, and is easily affected by operating habits and on-site noise, resulting in large fluctuations in lithology identification results. Although laboratory sampling and analysis can provide relatively accurate geological data, its testing cycle is long and the cost is high, making it difficult to meet the needs of real-time lithology identification during rapid tunnel excavation. Summary of the Invention

[0003] In view of this, the present invention proposes a real-time tunnel lithology identification method based on clustering algorithm, which aims to solve the problems of poor real-time performance, insufficient accuracy and low clustering stability in the current technology.

[0004] This invention proposes a real-time tunnel lithology identification method based on a clustering algorithm, comprising: A multidimensional drilling dataset is constructed based on the real-time collection of six key parameters from the rock drilling rig: propulsion speed, impact pressure, propulsion pressure, rotation pressure, water pressure, and water flow rate. Multidimensional drilling data is preprocessed using sliding window filtering and outlier removal algorithms to obtain a standardized dataset; The standardized dataset is mapped to a high-dimensional feature space to generate multi-dimensional feature vectors. An improved K-means++ clustering algorithm is then introduced to perform unsupervised classification of the multi-dimensional feature vectors to obtain the initial lithology classification. The centroid distribution is dynamically adjusted based on the initial lithological classification results, and the objective function is iteratively optimized until convergence, generating the final lithological classification. The final lithology classification results are linked with the geological database to output lithology identification results and guide the adjustment of construction parameters.

[0005] Furthermore, when preprocessing the multidimensional drilling data based on sliding window filtering and outlier removal algorithms to obtain a standardized dataset, the process includes: Multidimensional drilling data is smoothed using a sliding window filtering algorithm to remove high-frequency noise interference. The smoothed data is then processed to remove outliers based on a preset outlier detection threshold, resulting in preliminary cleaned data. The Z-score-based standardization method normalizes the initially cleaned data to obtain a standardized dataset.

[0006] Furthermore, when mapping the standardized dataset to a high-dimensional feature space to generate multi-dimensional feature vectors, the following steps are included: Principal component analysis (PCA) algorithm is used to reduce the dimensionality of standardized datasets in order to extract the main feature components. The main feature components are weighted and combined according to the preset feature weight allocation rules to generate a multidimensional feature vector; The kernel function mapping method projects multidimensional feature vectors into a high-dimensional feature space to enhance the ability to express nonlinear relationships between features.

[0007] Furthermore, when generating a multidimensional feature vector by weighting and combining the main feature components according to a preset feature weight allocation rule, the process includes: Initial weight values ​​are set according to the importance of each parameter to lithology identification, and the initial weight values ​​are optimized and adjusted based on a genetic algorithm to determine the weight allocation scheme. The main feature components are weighted and combined based on a weight allocation scheme to determine the multidimensional feature vector.

[0008] Furthermore, when projecting multidimensional feature vectors onto a high-dimensional feature space based on kernel function mapping methods to enhance the expressive power of nonlinear relationships between features, this includes: The radial basis function kernel is selected as the mapping tool, and the multidimensional feature vector is nonlinearly transformed according to the preset kernel parameters to determine the high-dimensional feature representation. Based on the high-dimensional feature representation, the complex correlation characteristics between features are enhanced, and the clustering analysis results are determined.

[0009] Furthermore, an improved K-means++ clustering algorithm is introduced to perform unsupervised classification of multidimensional feature vectors to obtain the initial lithology classification, including: Choose any sample point as the initial centroid, and use the square of the minimum distance from each sample point to the selected centroid as the probability distribution to select new centroids in turn, until K initial centroids are selected; Preliminary classification of multidimensional feature vectors is performed based on the initial centroid to obtain the initial lithology classification.

[0010] Furthermore, the centroid distribution is dynamically adjusted based on the initial lithology classification results, and iterative optimization is performed using the objective function until convergence. The final lithology classification includes: Based on the initial lithological classification results, the coordinates of the center point of each category are determined to update the centroid distribution; The multidimensional feature vectors are reclassified based on the updated centroid distribution to obtain new lithological category classifications; The new lithology classification results are evaluated based on a pre-defined objective function to obtain the loss value; Determine whether the loss value is lower than a preset convergence threshold, where: If the loss value is not lower than the convergence threshold, repeat the above steps until the loss value is lower than the convergence threshold to generate the final lithology classification.

[0011] Furthermore, when evaluating the new lithological classification results according to the preset objective function to obtain the loss value, the following steps are included: Obtain the sum of squared distances from each sample point within a category to the centroid to obtain the intra-class distance; Obtain the sum of squared distances between centroids of different classes to obtain the inter-class distance; The objective function is defined based on the ratio of intra-class distance to inter-class distance to obtain the loss value.

[0012] Furthermore, when linking the final lithology classification results with the geological database to output lithology identification results and guide the adjustment of construction parameters, this includes: Based on the final lithology classification results, the corresponding lithology attribute information is matched from the geological database to obtain the lithology identification results; The lithology identification results are correlated with real-time drilling parameters to generate suggestions for adjusting construction parameters; The visualization module displays the lithology identification results and construction parameter adjustment suggestions in real time to assist in construction decision-making.

[0013] Furthermore, the visualization module displays the lithology identification results and construction parameter adjustment suggestions in real time to assist construction decision-making, including: Spatial distribution display of lithology identification results based on 3D visualization technology; Trend analysis of construction parameter adjustment suggestions based on dynamic curves; The interactive interface enables real-time linkage between lithology identification results and construction parameter adjustment suggestions.

[0014] Compared with existing technologies, the advantages of this invention are as follows: By collecting six key parameters in real time—propulsion speed, impact pressure, propulsion pressure, rotation pressure, water pressure, and water flow rate—from a drilling rig, a multi-dimensional drilling dataset is constructed, comprehensively reflecting the mechanical response and formation characteristics during the drilling process. This multi-dimensional parameter fusion method can effectively capture subtle differences between different rock strata, providing a richer feature basis for subsequent lithology identification and significantly improving the accuracy and robustness of lithology classification. Secondly, the original drilling data is preprocessed using sliding window filtering and outlier removal algorithms, effectively reducing data bias caused by differences in operator experience, equipment fluctuations, and random noise. The standardized dataset has higher stability and consistency, thereby improving the convergence efficiency and computational reliability of the clustering algorithm and providing high-quality data input for real-time identification. Thirdly, automatic lithology classification is achieved through an improved K-means++ clustering algorithm. This algorithm introduces a distance-weighted random selection mechanism during centroid initialization, making the initial centroids more evenly distributed in the feature space, effectively avoiding the problem of traditional K-means algorithms easily getting trapped in local optima. By further combining dynamic centroid adjustment and iterative optimization of the objective function, the convergence speed and stability of the clustering results are improved, thus ensuring the accuracy and repeatability of lithology identification results. Finally, the final clustering results are linked with the geological database for output, enabling real-time visualization of lithology identification results and dynamic guidance for construction parameters. The system can automatically adjust drilling and blasting parameters or support schemes based on the identified lithology categories, significantly improving the intelligence and safety of the construction process. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a real-time tunnel lithology identification method based on a clustering algorithm, provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating a real-time tunnel lithology identification method based on a clustering algorithm, provided in an embodiment of the present invention. Detailed Implementation

[0016] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] like Figures 1-2 As shown in some embodiments of this application, this embodiment provides a real-time tunnel lithology identification method based on a clustering algorithm, including: Step S100: Based on the real-time acquisition of six key parameters by the rock drilling rig, namely propulsion speed, impact pressure, propulsion pressure, rotation pressure, water pressure and water flow rate, a multi-dimensional drilling dataset is constructed.

[0018] Step S200: Preprocess the multidimensional drilling data based on sliding window filtering and outlier removal algorithms to obtain a standardized dataset.

[0019] Specifically, when preprocessing multidimensional drilling data based on sliding window filtering and outlier removal algorithms to obtain a standardized dataset, the process includes: smoothing the multidimensional drilling data based on the sliding window filtering algorithm to remove high-frequency noise interference; removing outliers from the smoothed data according to a preset outlier detection threshold to obtain preliminary cleaned data; and normalizing the preliminary cleaned data based on the Z-score standardization method to obtain a standardized dataset.

[0020] Understandably, a multi-stage data preprocessing mechanism is used to suppress noise, remove outliers, and standardize values ​​for the multidimensional parameters collected during drilling, thereby improving the stability and feature expression accuracy of subsequent clustering algorithms and ensuring the reliability and consistency of lithology identification results. First, a sliding window filtering algorithm is used to smooth the original multidimensional drilling data obtained during drilling. This algorithm effectively reduces high-frequency noise interference and preserves the low-frequency variation trend during drilling by setting a fixed-length sliding window on the time series and calculating the mean or weighted average of the data within the window. Due to factors such as equipment vibration, instantaneous impact, and signal sampling errors during drilling, the original data often exhibits fluctuations and dispersion. Sliding window filtering can remove local disturbances while maintaining the overall variation characteristics, making the data more continuous and stable. Second, for the smoothed data, an outlier identification and cleaning algorithm based on a preset outlier detection threshold is used. This process analyzes the statistical distribution characteristics of each parameter (such as mean, standard deviation, extreme value range, etc.). When a data point deviates from the normal range by more than a set threshold, it is considered outlier data and removed. This step effectively eliminates abnormal fluctuations caused by factors such as improper operation, drill string jamming, and sensor transient errors, thus avoiding their interference with subsequent clustering analysis. Finally, a Z-score-based standardization method is used to normalize the initially cleaned data. This method calculates the difference between each data point and the overall mean and divides it by the standard deviation, transforming the range of different parameters to the same scale (usually a distribution with a mean of 0 and a standard deviation of 1). Since the dimensions and orders of magnitude of different drilling parameters differ significantly, directly inputting them into the clustering model can lead to high-amplitude features dominating the clustering results. Z-score standardization eliminates this scale difference, giving each dimension of features equal weight in the clustering analysis, thereby improving the accuracy of distance measurements between samples in the feature space.

[0021] Step S300: Map the standardized dataset to a high-dimensional feature space to generate multi-dimensional feature vectors, and introduce an improved K-means++ clustering algorithm to perform unsupervised classification of the multi-dimensional feature vectors to obtain the initial lithology category classification.

[0022] Specifically, when mapping a standardized dataset to a high-dimensional feature space to generate multi-dimensional feature vectors, the process includes: performing dimensionality reduction on the standardized dataset based on principal component analysis to extract the main feature components; weighting and combining the main feature components according to a preset feature weight allocation rule to generate multi-dimensional feature vectors; and projecting the multi-dimensional feature vectors to the high-dimensional feature space based on a kernel function mapping method to enhance the ability to express nonlinear relationships between features.

[0023] Specifically, when generating a multidimensional feature vector by weighting the main feature components according to the preset feature weight allocation rules, the process includes: setting initial weight values ​​based on the importance of each parameter to lithology identification, optimizing and adjusting the initial weight values ​​based on a genetic algorithm, and determining the weight allocation scheme; and weighting the main feature components based on the weight allocation scheme to determine the multidimensional feature vector.

[0024] Specifically, when projecting multidimensional feature vectors onto a high-dimensional feature space based on the kernel function mapping method to enhance the ability to express nonlinear relationships between features, the method includes: selecting a radial basis kernel function as a mapping tool, and performing a nonlinear transformation on the multidimensional feature vectors according to preset kernel parameters to determine the high-dimensional feature representation; and enhancing the complex correlation characteristics between features based on the high-dimensional feature representation to determine the clustering analysis results.

[0025] Specifically, when introducing an improved K-means++ clustering algorithm to perform unsupervised classification of multidimensional feature vectors to obtain the initial lithology classification, the following steps are taken: randomly select a sample point as the initial centroid, and select new centroid points in sequence according to the probability distribution of the minimum squared distance from each sample point to the selected centroid, until K initial centroids are selected; perform preliminary classification of multidimensional feature vectors based on the initial centroids to obtain the initial lithology classification.

[0026] Understandably, by integrating feature dimensionality reduction, weight optimization, kernel space mapping, and improved clustering algorithms, efficient feature extraction and refined clustering analysis of multidimensional drilling data can be achieved, thereby accurately identifying different lithological categories. This technical approach comprehensively utilizes linear and nonlinear feature processing methods to fully explore the potential correlations between drilling parameters, improving the resolution and algorithm stability of lithology identification. First, principal component analysis (PCA) is used to reduce the dimensionality of the standardized multidimensional drilling data. PCA extracts the main feature components that maximize the retention of the original data variance by calculating the covariance matrix between each parameter, thereby removing redundant and noise information and reducing data dimensionality. This process not only reduces computational complexity but also highlights the key features most relevant to lithological changes, enabling the subsequent clustering model to focus on the parameter dimensions that contribute the most to stratum identification. Second, in the feature combination stage, a weight optimization mechanism based on a genetic algorithm is introduced. Different drilling parameters have different importance for lithology determination; for example, impact pressure and drilling speed are more sensitive to the identification of hard rock strata, while water flow and water pressure are more representative in loose strata. To address this, the system sets initial weight values ​​based on the importance of each parameter and performs global optimization using a genetic algorithm. The clustering effect of the weight combination is evaluated through a fitness function, gradually evolving to the optimal weight allocation scheme. This method overcomes the subjectivity of manual weight setting, adaptively determining the parameter contribution based on data characteristics, thereby generating more discriminative multidimensional feature vectors. Then, a kernel function mapping method is used to project the multidimensional feature vectors onto a high-dimensional feature space to enhance the ability to express nonlinear relationships between features. By selecting the radial basis function (RBF kernel) as the mapping tool, the originally linearly inseparable feature relationships are transformed into linearly separable forms in a high-dimensional space, enabling the clustering algorithm to identify subtle differences between complex rock strata. This nonlinear mapping process effectively captures the nonlinear coupling features hidden in drilling data, such as the coordinated change patterns between pressure and flow, thereby improving the discrimination accuracy and robustness of clustering. Finally, an improved K-means++ clustering algorithm is introduced to perform unsupervised classification of the high-dimensional feature vectors. Traditional K-means algorithms are prone to getting trapped in local optima due to the random selection of initial centroids. The improved K-means++, however, introduces a probability distribution selection mechanism based on squared distance, making the initial centroids more evenly distributed in the feature space and effectively avoiding cluster bias. Subsequently, iterative updates are performed to gradually optimize the centroid positions, resulting in more stable clustering results and clearer classification boundaries, thus yielding the initial lithology classification results.

[0027] Furthermore, a core limitation of the traditional K-means algorithm lies in its completely random selection of initial centroids. While random initialization is simple, it easily leads to an overly concentrated distribution of centroids, causing multiple centroids to fall into high-density regions of the same cluster. This results in some true structural feature points not being covered, leading to problems such as narrowed convergence paths, unbalanced cluster boundaries, and high sensitivity of clustering results to initial points during iterative optimization. Ultimately, the algorithm often gets trapped in local optima, manifesting as weak differences between clusters, uneven cluster density, and poor cluster stability. To address the uncertainty caused by initial randomness, traditional K-means++ introduces a "probability selection based on squared distance" mechanism. After selecting the first centroid, it constructs a distribution based on the squared shortest distance between each point and the existing set of centroids, making points farther away more likely to become new centroids. In this way, K-means++ can ensure a more uniform distribution of initial centroids in the feature space, effectively avoiding multiple centroids falling into the same cluster region, thus significantly reducing the probability of getting trapped in local optima. However, traditional K-means++ can still suffer from initialization bias when dealing with high-dimensional sparse data, irregularly shaped data distributions, or large differences in cluster density. This bias can arise due to the reliance on a single distance metric or the failure to consider local density factors in the probability distribution. Consequently, a few centroids may be selected in low-density regions, weakening the overall clustering effect. To address these shortcomings, improved K-means++ typically incorporates more statistical or geometric information on top of the original probabilistic mechanism. For example, it may introduce local density estimation to avoid selecting centroids near noisy or marginal data points; add feature weights or similarity discrimination mechanisms to make the initial centroids more consistent with the inherent structure of the data; or utilize methods such as maximizing the minimum distance or weighted probability functions to ensure that the distribution of centroids is both uniform and matches the true clustering of the data clusters. The improved K-means++ can more accurately cover each potential clustering region, making its initial centroids more robust and representative of the data distribution, thus demonstrating higher consistency and faster convergence speed in multiple experiments. Overall, the randomness of traditional K-means leads to unstable clustering results. Traditional K-means++ effectively improves this problem by using the squared distance probability, thus improving the quality of cluster initialization. The improved K-means++ further enhances the adaptability to complex distributions, making the cluster boundaries more reasonable, the results more stable, and significantly reducing the interference of noise points on the initialization, thereby demonstrating higher accuracy and robustness in practical engineering applications.

[0028] Step S400: Dynamically adjust the centroid distribution based on the initial lithology classification results, and iteratively optimize the objective function until convergence, generating the final lithology classification.

[0029] Specifically, the process of dynamically adjusting the centroid distribution based on the initial lithology classification results and iteratively optimizing it in conjunction with the objective function until convergence, and generating the final lithology classification, includes: determining the coordinates of the center point of each category based on the initial lithology classification results to update the centroid distribution; reclassifying the multidimensional feature vectors based on the updated centroid distribution to obtain new lithology classifications; evaluating the new lithology classification results according to the preset objective function to obtain a loss value; and determining whether the loss value is lower than a preset convergence threshold. If the loss value is not lower than the convergence threshold, the above steps are repeated until the loss value is lower than the convergence threshold, generating the final lithology classification.

[0030] Specifically, when evaluating the new lithology classification results according to the preset objective function to obtain the loss value, the process includes: obtaining the sum of squared distances from sample points within each category to the centroid to obtain the intra-class distance; obtaining the sum of squared distances between centroids of different categories to obtain the inter-class distance; and defining the objective function based on the ratio of intra-class distance to inter-class distance to obtain the loss value.

[0031] Understandably, by using dynamic centroid updates and an iterative optimization mechanism based on an objective function, the system achieves adaptive refinement and convergence control of the lithology classification results, thereby obtaining a stable, accurate, and physically consistent final clustering result. This process is essentially a dynamic iterative process based on optimization criteria, continuously minimizing the clustering loss function to accurately characterize the lithology distribution pattern in the sample space. First, based on the lithology classification results obtained in the initial clustering stage, the system calculates the coordinates of the centroid point of each category and updates the centroid distribution. The centroid update is obtained by averaging the coordinates of all sample points within each cluster category. This process allows the new centroid to more accurately reflect the true centroid position of the category in the high-dimensional feature space. As the centroid position is adjusted, the cluster boundary is also optimized, providing a new reference benchmark for the next round of sample reclassification. Second, based on the new centroid distribution, the system reclassifies the multidimensional feature vectors, calculates the Euclidean distance between the sample and each centroid, and assigns the sample to the category corresponding to the nearest centroid. This process achieves adaptive evolution of the clustering structure, enabling the clustering model to gradually approach the optimal segmentation state. Meanwhile, to quantitatively evaluate the effectiveness of clustering optimization, an objective function (loss function) is introduced as an evaluation metric for clustering performance. The objective function is constructed based on the principle of "intra-cluster compactness and inter-cluster separation," and its mathematical expression is the ratio of intra-cluster distance to inter-cluster distance. Specifically, intra-cluster distance is obtained by calculating the sum of squared distances from sample points in each category to their centroids, reflecting the degree of clustering of samples within the same category; inter-cluster distance is obtained by calculating the sum of squared distances between the centroids of different categories, measuring the discriminative power between different categories. By taking the ratio of these two as the loss value, the intra-cluster variance can be minimized and the inter-cluster distance maximized during the optimization process, thus achieving a comprehensive balance in clustering performance. Finally, the system determines whether the loss value calculated by the objective function is lower than a preset convergence threshold. When the loss value is still greater than the threshold, it indicates that the clustering results are not yet stable and the iterative process of "centroid update - sample reclassification - objective function evaluation" needs to continue. When the loss value is lower than the threshold, it indicates that the clustering structure is becoming stable, that is, the samples within the cluster are highly concentrated and the differences between the categories are obvious. At this time, the iteration terminates and the final lithology classification result is output.

[0032] Step S500: Link the final lithology classification results with the geological database, output lithology identification results, and guide the adjustment of construction parameters.

[0033] Specifically, when linking the final lithology classification results with the geological database to output lithology identification results and guide the adjustment of construction parameters, the process includes: matching the corresponding lithology attribute information from the geological database based on the final lithology classification results to obtain lithology identification results; performing correlation analysis between the lithology identification results and real-time drilling parameters to generate suggestions for adjusting construction parameters; and displaying the lithology identification results and suggestions for adjusting construction parameters in real time based on the visualization module to assist in construction decision-making.

[0034] Specifically, the visualization module displays the lithology identification results and construction parameter adjustment suggestions in real time to assist construction decisions. This includes: displaying the spatial distribution of lithology identification results based on 3D visualization technology; performing trend analysis on construction parameter adjustment suggestions based on dynamic curves; and achieving real-time linkage between lithology identification results and construction parameter adjustment suggestions through an interactive interface.

[0035] Understandably, by linking clustering results with geological databases, using parameter correlation analysis models, and employing dynamic visualization technology, structured output of lithology identification results and intelligent assistance for construction decisions can be achieved, thus forming a data-driven, integrated closed-loop control system encompassing lithology, parameters, and construction. This principle deeply integrates algorithmic identification results with engineering applications, enabling not only lithology determination but also proactive guidance for construction parameter optimization and risk control. First, through the linkage and matching of clustering results with geological databases, a mapping from cluster categories to specific lithological attributes is achieved. Based on the final lithology category classification results, lithological information corresponding to that category (such as rock name, hardness coefficient, water content, integrity index, etc.) is retrieved from the geological database, transforming the abstract categories of unsupervised clustering into actual lithology types with geological significance. The core principle of this step lies in establishing a one-to-one correspondence between "cluster labels—geological attributes," making the data-driven classification results interpretable and providing a basis for subsequent adjustments to construction parameters. Second, based on correlation analysis algorithms, lithology identification results are dynamically coupled and analyzed with real-time drilling parameters. This process identifies the optimal range of construction parameters by calculating the statistical regularities of parameters such as drilling speed, impact pressure, slewing torque, and water flow under different lithological conditions. When real-time drilling parameters deviate from this range, it automatically generates suggestions for adjusting construction parameters, such as reducing impact frequency, optimizing drilling speed, or adjusting cooling water pressure. Its technical principle lies in using data correlation analysis models (such as Pearson correlation coefficient and mutual information analysis) to uncover the quantitative relationship between lithology and construction behavior, thereby achieving intelligent parameter optimization and risk warning. Furthermore, it introduces 3D visualization and dynamic interactive technology to achieve real-time visualization of lithology identification results and parameter adjustment suggestions. Through the 3D visualization module, the identified lithology categories are displayed in layers in the tunnel space model using different colors or textures, intuitively presenting the spatial distribution characteristics of different rock layers. Simultaneously, dynamic curves are used to display the changing trends of construction parameters and adjustment suggestions, allowing operators to intuitively grasp the response characteristics of the drilling process. Combined with an interactive interface, users can click, zoom, or select different sections in real time to view local lithology information and corresponding construction optimization suggestions, achieving interactive linkage between data and on-site operations. Finally, this visualization and linkage mechanism forms a closed-loop feedback loop of identification, analysis, and decision-making. It not only outputs lithological identification results but also adjusts construction strategies in real time based on the identification conclusions, achieving closed-loop optimization from data analysis to engineering control. This data-driven intelligent decision-making mechanism effectively improves the safety, accuracy, and automation level of tunnel construction, reduces human judgment errors and unnecessary trial operations, and promotes the deep integration of intelligent tunnel equipment and geological information.

[0036] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principle of this invention is further explained below in conjunction with a specific application scenario.

[0037] In tunnel construction, drilling rigs are widely used as core equipment in drilling operations. Firstly, the drilling rig's sensors are positioned near the drill bit to collect six key parameters in real time: propulsion speed, impact pressure, propulsion pressure, rotational pressure, water pressure, and water flow rate. These parameters are transmitted to the central processing unit via wired or wireless communication modules, ensuring the real-time nature and integrity of the data. For example, in a certain construction scenario, the drilling rig collects data at a frequency of 10 times per second and transmits the collected data to the central processing unit for subsequent analysis. To ensure the accuracy of data acquisition, the sensors need to be calibrated regularly, and the sampling frequency adjusted according to the construction environment. For example, when working in hard rock formations, the sampling frequency can be appropriately increased to capture more subtle changes; while when working in soft rock formations, the sampling frequency can be appropriately decreased to reduce data redundancy.

[0038] The collected multidimensional drilling data then enters the preprocessing stage. A sliding window filtering algorithm is applied to smooth the raw data. The width of the sliding window is set to 5 seconds, and the data points within the window generate new smoothed values ​​through mean calculation. For example, in one embodiment, when the collected impact pressure data generates high-frequency noise due to equipment vibration, the sliding window filtering algorithm can effectively reduce noise interference and retain the core trend of the data. Subsequently, an outlier removal algorithm based on statistical principles is applied to the smoothed data. This algorithm removes outliers by calculating the deviation of each data point from its neighboring points and combining it with a preset detection threshold. For example, when the deviation of a data point exceeds three times the standard deviation, the point is considered an outlier and removed. Finally, a Z-score-based standardization method is used to normalize the initially cleaned data. Z-score standardization eliminates the interference of differences in the dimensions of different parameters on subsequent analysis by normalizing the mean and variance of the data. After the above steps, the obtained standardized dataset not only retains the core information of the original data but also significantly improves the usability and reliability of the data.

[0039] The standardized dataset is then mapped to a high-dimensional feature space to generate multi-dimensional feature vectors. This process begins with dimensionality reduction of the standardized dataset using Principal Component Analysis (PCA). PCA extracts the principal feature components that best reflect data variability by performing eigenvalue decomposition on the covariance matrix. For example, in one embodiment, the original six-dimensional dataset is reduced to three dimensions, retaining over 95% of the data variability. Next, the principal feature components are weighted according to a preset feature weighting rule. Initial weight values ​​are set based on the importance of each parameter to lithology identification; for example, the initial weight value for impact pressure is set to 0.3, and for thrusting pressure to 0.25, with the weights of other parameters decreasing sequentially. Subsequently, a genetic algorithm is used to optimize and adjust the initial weight values. The genetic algorithm gradually optimizes the weight allocation scheme by encoding, crossover, and mutation operations on the weight values. In one embodiment, after 100 generations of evolution, the optimal weight allocation scheme is determined, with the final weight value for impact pressure adjusted to 0.35 and for thrusting pressure to 0.28. After weighting and combining the main feature components based on the optimal weight allocation scheme, the resulting multidimensional feature vector can comprehensively characterize the mechanical response characteristics during the drilling process.

[0040] The generated multidimensional feature vectors are then projected onto a high-dimensional feature space using a kernel function mapping method to enhance the ability to express nonlinear relationships between features. In this embodiment, the radial basis function (RBF) kernel is selected as the mapping tool. The RBF kernel significantly enhances the complex correlation characteristics between features by performing a nonlinear transformation on the multidimensional feature vectors. For example, in one embodiment, the three-dimensional feature vectors are mapped to a five-dimensional feature space, and the kernel parameter σ is set to 0.5. The high-dimensional feature representation generated after kernel function mapping not only accurately reflects the differences between different rock strata but also significantly improves the classification accuracy and stability of cluster analysis.

[0041] The high-dimensional feature representations are then input into an improved K-means++ clustering algorithm for unsupervised classification. The improved K-means++ algorithm employs a distance-based probability distribution strategy in the initial centroid selection process. Specifically, after randomly selecting a sample point as the initial centroid, new centroids are selected sequentially based on the probability distribution of the minimum squared distance from each sample point to the selected centroid, until K initial centroids are selected. For example, in one embodiment, the value of K is set to 5, ensuring a uniform distribution of centroids in the feature space. After preliminary classification of the multidimensional feature vectors based on the initial centroids, the generated initial lithology classification provides a reliable starting point for subsequent iterative optimization.

[0042] The initial lithology classification results are then dynamically adjusted by adjusting the centroid distribution and iteratively optimizing the objective function until convergence is achieved, generating the final lithology classification. Specifically, the centroid distribution is updated by calculating the coordinates of the center point of each category based on the initial lithology classification results. The multidimensional feature vectors are then reclassified based on the updated centroid distribution to obtain new lithology classifications. The objective function quantifies the quality of the clustering results by comprehensively evaluating intra-class and inter-class distances. For example, in one embodiment, the intra-class distance is defined as the sum of the squared distances from sample points within each category to the centroid, and the inter-class distance is defined as the sum of the squared distances between centroids of different categories. The objective function calculates the loss value by measuring the ratio of intra-class to inter-class distances. When the loss value is not lower than a preset convergence threshold, the above steps are repeated until the loss value is lower than the convergence threshold. For example, in one embodiment, the convergence threshold is set to 0.01, and after 15 iterations, the loss value drops to 0.008, at which point the generated final lithology classification has high classification accuracy and stability.

[0043] The final lithology classification results are then linked with the geological database to output lithology identification results and guide the adjustment of construction parameters. The geological database provides authoritative attribute information support for the lithology identification results through the accumulation and organization of historical lithology data. For example, in one embodiment, the geological database contains mechanical response characteristic data for over 1000 lithology samples. After matching the corresponding lithology attribute information from the geological database based on the final lithology classification results, the resulting lithology identification results are correlated with real-time drilling parameters. This correlation analysis generates scientifically sound and reasonable suggestions for adjusting construction parameters through cross-comparison of the lithology identification results and real-time drilling parameters. For example, in one embodiment, when the identification result is a hard rock layer, it is recommended to increase the impact pressure and reduce the advance speed to optimize drilling efficiency. The visualization module uses 3D visualization technology to display the spatial distribution of the lithology identification results and performs trend analysis on the construction parameter adjustment suggestions through dynamic curves. The interactive interface, through real-time linkage, seamlessly connects the complex lithology identification results with the construction parameter adjustment suggestions, significantly improving decision-making efficiency and accuracy.

[0044] The above scenarios are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0045] In the above embodiments, six key parameters—propulsion speed, impact pressure, propulsion pressure, slewing pressure, water pressure, and water flow rate—are collected in real time by a drilling rig to construct a multi-dimensional drilling dataset, comprehensively reflecting the mechanical response and formation characteristics during the drilling process. This multi-dimensional parameter fusion method effectively captures subtle differences between different rock strata, providing a richer feature base for subsequent lithology identification and significantly improving the accuracy and robustness of lithology classification. Secondly, a sliding window filtering and outlier removal algorithm are used to preprocess the original drilling data, effectively reducing data bias caused by differences in operator experience, equipment fluctuations, and random noise. The standardized dataset has higher stability and consistency, thereby improving the convergence efficiency and computational reliability of the clustering algorithm and providing high-quality data input for real-time identification. Thirdly, an improved K-means++ clustering algorithm is used to achieve automatic lithology classification. This algorithm introduces a distance-weighted random selection mechanism during centroid initialization, making the initial centroids more evenly distributed in the feature space and effectively avoiding the problem of traditional K-means algorithms easily getting trapped in local optima. By further combining dynamic centroid adjustment and iterative optimization of the objective function, the convergence speed and stability of the clustering results are improved, thus ensuring the accuracy and repeatability of lithology identification results. Finally, the final clustering results are linked with the geological database for output, enabling real-time visualization of lithology identification results and dynamic guidance for construction parameters. The system can automatically adjust drilling and blasting parameters or support schemes based on the identified lithology categories, significantly improving the intelligence and safety of the construction process.

[0046] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0047] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0048] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0049] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for real-time identification of tunnel lithology based on clustering algorithm, characterized in that, include: A multidimensional drilling dataset is constructed based on the real-time collection of six key parameters from the rock drilling rig: propulsion speed, impact pressure, propulsion pressure, rotation pressure, water pressure, and water flow rate. Multidimensional drilling data is preprocessed using sliding window filtering and outlier removal algorithms to obtain a standardized dataset; The standardized dataset is mapped to a high-dimensional feature space to generate multi-dimensional feature vectors. An improved K-means++ clustering algorithm is then introduced to perform unsupervised classification of the multi-dimensional feature vectors to obtain the initial lithology classification. The centroid distribution is dynamically adjusted based on the initial lithological classification results, and the objective function is iteratively optimized until convergence, generating the final lithological classification. The final lithology classification results are linked with the geological database to output lithology identification results and guide the adjustment of construction parameters.

2. The real-time tunnel lithology identification method based on clustering algorithm as described in claim 1, characterized in that, When preprocessing multidimensional drilling data using sliding window filtering and outlier removal algorithms to obtain a standardized dataset, the process includes: Multidimensional drilling data is smoothed using a sliding window filtering algorithm to remove high-frequency noise interference. The smoothed data is then processed to remove outliers based on a preset outlier detection threshold, resulting in preliminary cleaned data. The Z-score-based standardization method normalizes the initially cleaned data to obtain a standardized dataset.

3. The real-time tunnel lithology identification method based on clustering algorithm as described in claim 1, characterized in that, Mapping a standardized dataset to a high-dimensional feature space to generate multi-dimensional feature vectors includes: Principal component analysis (PCA) algorithm is used to reduce the dimensionality of standardized datasets in order to extract the main feature components. The main feature components are weighted and combined according to the preset feature weight allocation rules to generate a multidimensional feature vector; The kernel function mapping method projects multidimensional feature vectors into a high-dimensional feature space to enhance the ability to express nonlinear relationships between features.

4. The real-time tunnel lithology identification method based on clustering algorithm as described in claim 3, characterized in that, When generating a multidimensional feature vector by weighting and combining the main feature components according to a preset feature weight allocation rule, the process includes: Initial weight values ​​are set according to the importance of each parameter to lithology identification, and the initial weight values ​​are optimized and adjusted based on a genetic algorithm to determine the weight allocation scheme. The main feature components are weighted and combined based on a weight allocation scheme to determine the multidimensional feature vector.

5. The real-time tunnel lithology identification method based on clustering algorithm as described in claim 3, characterized in that, When projecting multidimensional feature vectors onto a high-dimensional feature space using kernel function mapping methods to enhance the expressive power of nonlinear relationships between features, the following methods are employed: The radial basis function kernel is selected as the mapping tool, and the multidimensional feature vector is nonlinearly transformed according to the preset kernel parameters to determine the high-dimensional feature representation. Based on the high-dimensional feature representation, the complex correlation characteristics between features are enhanced, and the clustering analysis results are determined.

6. The method for real-time identification of tunnel lithology based on clustering algorithm as described in claim 1, characterized in that, An improved K-means++ clustering algorithm is introduced to perform unsupervised classification of multidimensional feature vectors to obtain the initial lithology classification, including: Choose any sample point as the initial centroid, and use the square of the minimum distance from each sample point to the selected centroid as the probability distribution to select new centroids in turn, until K initial centroids are selected; Preliminary classification of multidimensional feature vectors is performed based on the initial centroid to obtain the initial lithology classification.

7. The method for real-time identification of tunnel lithology based on clustering algorithm as described in claim 1, characterized in that, The centroid distribution is dynamically adjusted based on the initial lithological classification results, and iterative optimization is performed using the objective function until convergence. The final lithological classification includes: Based on the initial lithological classification results, the coordinates of the center point of each category are determined to update the centroid distribution; The multidimensional feature vectors are reclassified based on the updated centroid distribution to obtain new lithological category classifications; The new lithology classification results are evaluated based on a pre-defined objective function to obtain the loss value; Determine whether the loss value is lower than a preset convergence threshold, where: If the loss value is not lower than the convergence threshold, repeat the above steps until the loss value is lower than the convergence threshold to generate the final lithology classification.

8. The real-time tunnel lithology identification method based on clustering algorithm as described in claim 7, characterized in that, When evaluating the new lithological classification results based on the preset objective function to obtain the loss value, the following are included: Obtain the sum of squared distances from each sample point within a category to the centroid to obtain the intra-class distance; Obtain the sum of squared distances between centroids of different classes to obtain the inter-class distance; The objective function is defined based on the ratio of intra-class distance to inter-class distance to obtain the loss value.

9. The method for real-time identification of tunnel lithology based on clustering algorithm as described in claim 1, characterized in that, When linking the final lithology classification results with the geological database to output lithology identification results and guide the adjustment of construction parameters, the following should be included: Based on the final lithology classification results, the corresponding lithology attribute information is matched from the geological database to obtain the lithology identification results; The lithology identification results are correlated with real-time drilling parameters to generate suggestions for adjusting construction parameters; The visualization module displays the lithology identification results and construction parameter adjustment suggestions in real time to assist in construction decision-making.

10. The method for real-time identification of tunnel lithology based on clustering algorithm as described in claim 9, characterized in that, The visualization module displays lithology identification results and construction parameter adjustment suggestions in real time to assist construction decision-making, including: Spatial distribution display of lithology identification results based on 3D visualization technology; Trend analysis of construction parameter adjustment suggestions based on dynamic curves; The interactive interface enables real-time linkage between lithology identification results and construction parameter adjustment suggestions.

Images